More potential for memristors

[unixronin]

HP has discovered that memristors have more potential than they previously realized.  They're apparently talking about the potential of building machines in which, instead of having a dedicated CPU with separate main memory, processing could be distributed throughout the memory and performed directly on the data in individual memory cells without having to move it to CPU registers first.  If I understand correctly, they're implying that individual memory cells could be "smart".  Since memristors are non-volatile, this opens up the possibility of a homogeneous-architecture machine that has no distinction between memory, processor and mass storage — just a single unified pool of non-volatile solid-state storage, all of which is capable of processing its own contents.

I think we'd have to come up with a new term of art for such architectures, a superlative of "massively parallel".  Perhaps we might refer to such machines as fully parallel architectures.  In such an architecture, if you needed more processing power, you could just add more memory, for perhaps a near-linear increase in throughput.

Such a fully-parallel machine would have astounding capabilities, many of them heretofore thought to be the sole province of as-yet-unbuilt quantum computers.  For example, imagine the speed of a complex query against a huge database on a machine that has the capability to examine every record simultaneously — or the potential for science or mathematics of a machine on which performing a transform on a pair of one-million-by-one-million matrices is a single atomic operation.

There could be signicant impacts on cryptography, too, strengthening both crypto and cryptologic attacks.  On the one hand, the speed of complex new encryption algorithms could become much less important on a fully parallel machine that can encrypt or decrypt an entire dataset simultaneously.  On the other hand, the ability to simultaneously try vast swathes of a keyspace could hugely speed up brute-force attacks.  However, that might become much less important given new algorithms made to exploit the strengths of fully parallel architectures.

Perhaps most dramatically, imagine the potential for machine vision and related fields.  Fully parallel architectures might even enable us, for the first time, to simulate function comparable to that of organic brains in real time, which we can hypothesize might even lead to true artificial sentiences.

The C|Net article mentions that memristors could break Moore's Law.  That's not all they have the potential to break, by a long way.  This could turn out to be a revolutionary change in computer technology, not an evolutionary one.

Comments

[unixronin.livejournal.com]

I inferred that's what you meant. :) It's definitely going to bear watching. It'd be nice to see HP return to greatness, too.

[wyrdling.livejournal.com]

that would help wash out this bitter taste in the mouth they've given me over the past decade, for sure...